As artificial intelligence systems grow exponentially more powerful, the biggest challenge isn't computing anymore,it's moving data between chips fast enough without burning through electricity. GlobalFoundries, a major semiconductor manufacturer, just secured a $300 million federal award to solve this problem by shifting data centers from copper wiring to optical technology that transmits information using light.

Why Is Data Movement Becoming the Real Bottleneck in AI?

The scale of AI infrastructure being built right now is staggering. By the end of 2028, the world is expected to have roughly 200 million AI chips in operation, up from about 20 million today, according to research firm Epoch AI. That's a tenfold increase in less than three years. Each of these chips generates enormous amounts of heat and needs to exchange data with thousands of other chips constantly.

Traditional copper interconnects, which have been the standard for decades, are hitting their limits. They consume significant power just to move electrical signals between chips, and as data volumes explode, this becomes a major source of energy waste. Silicon photonics solves this by using light instead of electricity to transmit data, which is inherently more efficient at high speeds and over longer distances.

"Silicon photonics is essential to AI infrastructure. For a decade, the industry talked about the shift from copper to optical as something that was coming,today it is here, moving data at higher bandwidth and improved power efficiency as workloads grow more complex," said Tim Breen, CEO of GlobalFoundries.

Tim Breen, CEO of GlobalFoundries

What Exactly Is Silicon Photonics, and How Does It Work?

Silicon photonics is a technology that uses light waves to transmit data instead of electrical signals traveling through copper wires. Think of it like replacing a crowded highway with a fiber optic cable that can carry vastly more information simultaneously. The technology enables what the industry calls "co-packaged optics" (CPO) and "near-packaged optics" (NPO), which integrate optical components directly into or very close to computing chips.

GlobalFoundries' newly announced SCALE platform (Silicon Photonics Co-Packaged Advanced Light Engine) targets industry-leading performance at 400 gigabits per second while delivering a 5x increase in energy efficiency compared to current copper-based systems. For data centers consuming megawatts of power, a fivefold efficiency gain is transformative.

How Will This Technology Reshape Data Center Design?

The shift to optical interconnects requires changes across multiple layers of data center architecture. GlobalFoundries will develop new optical materials, advanced wafer technologies, and specialized packaging techniques to make this transition possible at scale. The company plans to leverage existing manufacturing facilities in Malta, New York, and Burlington, Vermont, to produce these components domestically.

This is significant because it keeps critical semiconductor manufacturing in the United States, reducing dependence on overseas suppliers during a period when AI infrastructure is becoming a geopolitical priority. The U.S. Department of Commerce is even taking a 1 percent equity stake in GlobalFoundries as part of the deal, ensuring the public benefits from the company's growth.

Industry leaders across the ecosystem are backing this transition. AMD, Broadcom, Cisco, Corning, and Lumentum all issued statements supporting the investment, recognizing that optical interconnects are essential for the next generation of AI systems.

Steps to Understanding Silicon Photonics in Data Centers

• Current Problem: Copper interconnects consume excessive power and create heat bottlenecks as AI chips exchange massive amounts of data, limiting how efficiently data centers can scale.
• Optical Solution: Silicon photonics uses light instead of electricity to transmit data, enabling higher bandwidth with lower power consumption and reduced heat generation.
• Manufacturing Scale: GlobalFoundries will use proven 3D hybrid bonding and advanced packaging techniques to manufacture optical components at high volume in U.S. facilities.
• Performance Target: The SCALE platform aims to deliver 400 gigabits per second of data throughput while consuming 80 percent less energy than current copper-based systems.

Why Is This Announcement Happening Right Now?

The timing reflects the urgency of the AI infrastructure boom. Hundreds of massive data centers are under construction across the United States and globally, with hundreds of millions of dollars being invested annually. By 2029, global AI infrastructure investment is forecast to exceed $1 trillion, according to market research firm IDC. At that scale, even small improvements in energy efficiency translate to billions of dollars in operational savings.

Data center operators are also facing mounting pressure from communities concerned about electricity consumption and environmental impact. A more efficient interconnect technology directly addresses these concerns by reducing the total power footprint of AI infrastructure. Additionally, as AI systems become more geopolitically sensitive, keeping advanced semiconductor manufacturing domestic is increasingly important to U.S. policymakers.

"The CHIPS R&D incentives will support a breakthrough in compute and communication networks moving past traditional copper bottlenecking to power next-generation AI," explained Bill Frauenhofer, Executive Director for Semiconductor Innovation and Investment at the Department of Commerce.

Bill Frauenhofer, Executive Director for Semiconductor Innovation and Investment at the Department of Commerce

What Does This Mean for the Future of AI Infrastructure?

Silicon photonics represents a fundamental shift in how data centers will be designed and operated. Rather than treating data movement as a secondary concern, it becomes a primary design consideration, much like compute power itself. Companies building the next generation of AI systems will need to account for optical interconnects from the ground up, not retrofit them later.

The technology also opens doors to new data center architectures that were previously impossible. With optical interconnects, chips can be distributed across larger physical distances while maintaining the low-latency, high-bandwidth connections that AI workloads demand. This flexibility could enable more efficient cooling, better power distribution, and novel approaches to scaling AI clusters.

For the broader AI industry, this investment signals that the infrastructure race is moving beyond just acquiring more chips. The companies that master efficient data movement will have a competitive advantage in building the largest, most cost-effective AI systems. As one industry executive noted, the bottleneck in AI infrastructure is shifting from compute to connectivity, and GlobalFoundries' $300 million federal award is designed to ensure that the United States leads in solving that challenge.